EdTech Research Evidence Standards Parents and Educators Should Demand
Schools and families need concrete standards to evaluate educational tools before adopting them.

Parents and educators are routinely asked to choose learning tools before those tools have earned the evidence that would justify choosing them. The market for children's educational technology has grown faster than the research meant to evaluate it. That leaves families and schools with a hard problem: not a shortage of options, but no clear way to tell which options actually work.
The pressure to adopt starts early and comes from every direction. District leaders face walls of apps, each promising proven results. Parents scroll app stores where every product claims to be grounded in research. Meanwhile, AI-based tools have pushed adoption even further ahead of evaluation. Most teachers and students were using AI in the 2024–25 school year, well before systematic research on its effects was complete.
The cost of choosing poorly is not neutral. Evidence on reading and math development makes clear that instructional gaps created early are harder to close over time. The practical fix is not to wait for perfect research, but to apply a concrete evaluation standard before adopting any tool. That standard is what this guide sets out.
What "evidence-based" actually means in reading instruction, and why it sets the bar
Reading instruction offers the clearest reference point for what a real evidence base looks like. Decades of converging scientific research sit behind what is now widely called the Science of Reading. This is not a single program or a philosophy. It is a large, interdisciplinary body of research from cognitive science, neuroscience, psychology, and education, and it converges on one conclusion: reading does not develop naturally and must be explicitly taught.
The research identifies five components as essential. The National Reading Panel established them: phonemic awareness, phonics, fluency, vocabulary, and comprehension. These have been replicated across multiple disciplines and languages, which matters for any claim to evidence. A product that omits one of them is not aligned with the research regardless of what its marketing page says.
The Simple View of Reading gives educators a structural model rather than just a list of activities. It frames comprehension as the product of decoding and language comprehension. That means both parts must work for reading to work. A child who can sound out words but cannot understand them will not comprehend. A child who understands language but cannot decode will not read. The model explains why each component exists and why skipping one breaks the whole.
The instructional implication is equally specific. Evidence-based structured literacy is explicit, sequential, systematic, and cumulative. You can ask whether instruction follows a planned order. These checks do not require a research degree.
What the research rules out is as important as what it endorses. The three-cueing system, prompting children to guess words using pictures, context, or initial sounds, lacks empirical support and is now explicitly excluded from evidence-based definitions in state law. Australian states, U.S. states, and other jurisdictions are mandating phonics screening and evidence-based curriculum. That regulatory signal tells you the research has cleared a high bar.
For a parent evaluating an app, the test is concrete. Ask whether its reading curriculum is explicit and sequential. Ask whether it covers all five components. Ask whether it avoids cueing strategies. Ask whether it can cite the specific research it rests on, not just the phrase "research-based." A product that passes those checks is grounded in something real.
How early math evidence works — and why the same logic applies
Early numeracy has its own strong longitudinal evidence base. A math product's claims deserve the same scrutiny as a reading product's claims. The research on how children develop number sense is clear, and it carries a warning for any tool that treats math as isolated skill-drilling.
Language and number development are intertwined from ages two and a half to three. General language and numeracy mutually predict each other's development from that window onward. That means a math tool that ignores language is misaligned with how the brain actually builds these capacities. A product that strips language out of early math instruction works against the developmental research.
Current informal settings, including many apps, are not producing optimal early numeracy gains. Parents have a straightforward evaluative question: does this math product demonstrate awareness of developmental research on early numeracy, and does it target skills at the window when they matter most? The same scrutiny applied to reading works here too.
What AI tutoring can genuinely do for young learners, and where the evidence stops
AI tutoring has a real and growing evidence base, but it is narrow and implementation-dependent. Meta-analytic evidence finds that AI-enabled adaptive learning systems can produce medium to large positive effects on learner cognitive outcomes. The strongest results come in procedural skill development, where the system adjusts difficulty and pacing to the individual learner. A child practicing number facts or decoding gets feedback at the right moment, with the next problem chosen to match what they just did. That is a genuine capability.
The same World Bank review contains a critical caution. An AI tool that adjusts difficulty while keeping the child doing the cognitive work teaches skill. The difference is in the design, not the technology.
AI cannot do some things. It cannot provide genuine emotional attunement. It cannot facilitate authentic peer collaboration. It cannot spark the kind of original creative thinking children develop through open-ended human interaction. Those limits matter most for young learners, whose development depends heavily on relationships and play.
For a parent or educator, the evaluative question is direct. Or is the AI doing the work while the child watches? The answer determines whether the tool builds skill or builds dependency.
Why "research-based" on a marketing page is almost never enough on its own
The phrase "research-based" has been diluted to meaninglessness. A product page can use that phrase while the product itself has never been studied. Understanding the specific ways evidence claims can be technically true and practically empty is essential.
One common red flag is citing a general body of research without specifying that the product itself has been studied. Alignment describes the design. It does not demonstrate the outcome.
Vendor-funded pilot studies are another red flag. A pilot run by the vendor is a starting point, not a conclusion. Treat it as preliminary.
Testimonials and case studies are not controlled research. A case study tells one story. Evidence tells many.
Then there is standards-as-evidence. Standards alignment describes what is taught, not whether children learn it. A product can be perfectly aligned to standards and still fail to produce gains.
Engagement theater is common. Time-on-app, stars earned, and streaks presented as evidence of learning are engagement metrics, not learning metrics. Enjoyment matters for motivation, but it is not achievement.
"Efficacy research in progress" deserves caution too. It is a reasonable statement for a new product, but not a justification for choosing it over one with completed research. For a child who is struggling, choosing a tool still being validated carries real cost.
Three non-negotiable questions belong in any serious evaluation. For reading specifically, the three-cueing system is a bright-line disqualifier. If a product instructs children to guess words from pictures, context, or initial sounds rather than decode, it contradicts the research base regardless of any other claim it makes.
What strong evidence actually looks like in practice, across a range of tools
Strong evidence in edtech has recognizable features. It is independent, replicable, population-specific, and connected to mechanisms of how children learn. Parents can learn to recognize these features without a research background.
For reading tools, strong evidence means the product has been studied specifically on the age group and skill level in question. Different ages, different skills, different evidence. A vendor claiming evidence from one population while marketing to another is a warning sign.
Diagnostic capability is a marker of quality. Evidence-based structured literacy requires continued progress monitoring to guide instruction and identify students who are struggling or falling behind grade level. A tool that does not track and respond to individual progress is not implementing the research, no matter what it claims.
State-level screening frameworks illustrate what rigorous evaluation looks like. Indiana's 2025–26 protocol evaluates universal screeners across multiple criteria, including type, purpose, and specific components related to dyslexia. That is a standard parents can reference when evaluating any progress-monitoring tool. Ask whether a product measures what matters, or only what is easy to measure.
For AI tools specifically, the evidence should show not just that outcomes improved, but that children were doing the cognitive work. Tools that supply answers do not. The mechanism matters as much as the outcome.
How to track whether a tool is actually working for your child
Choosing an evidence-based tool is necessary but not sufficient. A product can be well-designed and still fail for a particular child. Parents and educators need their own benchmarks to verify that a tool is producing gains for this child, not just for the children in a vendor's study.
For reading, formal screening tools provide measurable benchmarks independent of any app's internal reporting. Parents can ask schools to share this data or reference it when evaluating progress. The key is that the measure comes from outside the product's own dashboard.
Progress monitoring is a requirement, not a feature. A tool that does not surface progress data to parents is removing a key safeguard. Ask for it.
For math, the absence of a ceiling effect in early numeracy assessments is a signal. A child scoring well below what they are capable of on an external measure, despite regular app use, may not be reaching their potential. Optimal development is not guaranteed just because a child is using a tool regularly.
A practical rhythm works like this. If they diverge significantly, investigate why. A child who enjoys an app and uses it daily but shows flat scores on external assessments is a warning sign, not a success story.
Parents of children who are struggling or have IEPs have additional leverage. IDEA requires regular data collection and progress monitoring toward IEP goals. The IEP team is expected to make instructional adjustments when a student is not making adequate progress. Data from an app is testable evidence in an IEP conversation.
The five questions to ask before adopting any edtech product
A parent or educator who asks five specific questions before adopting any edtech product will be better protected than one who relies on marketing claims, ratings, or word of mouth alone. Each question follows directly from the evidence standards established above. Question one: What is the evidence, and who conducted it? Look for independent, peer-reviewed research on the specific product with the specific age group. Independent replication separates genuine evidence from marketing.
Question two: What is the curriculum grounded in? For reading, ask whether it covers all five components explicitly and sequentially, and whether it excludes cueing strategies. For math, ask whether it is aligned to developmental research on early numeracy, not just to standards. Curriculum grounding tells you what the product is built on.
Question three: Does the tool adapt to the individual child, or does it deliver the same content to every child? Genuine personalization means adjusting what is taught next, how it is taught, and when to move on, based on real-time response, not level placement. Delivering the same sequence to every child is not adaptation.
Question four: How does it protect a child's data? FERPA, COPPA, and GDPR compliance are minimum requirements, not differentiators. A vendor that cannot answer clearly is not protecting your child.
Question five: How will a parent know if it is working? It should surface progress data to parents in a form they can compare to external benchmarks. If the tool's only measure of success is its own engagement dashboard, that is not enough.
These questions apply equally to free and paid tools, AI and non-AI tools, apps and curricula. The evidence standard does not change based on the price point or the technology. A free app with no evidence is no better a choice than an expensive app with no evidence.
For parents of struggling readers, one additional filter applies. A child who is already behind cannot afford a tool that is still being validated. The cost of a wrong choice is measured in months of lost development, and those months do not come back.
Demanding evidence is not hostility to innovation. It is the condition that makes genuine innovation trustworthy. A product that has earned its evidence has earned the right to be chosen. One that has not can wait until it has.
Sources
- 8 science of reading strategies - Teach. Learn. Grow.
- Supporting fluency and comprehension using practices grounded in the science of reading - Teach. Learn. Grow.
- Early numeracy and mathematics development: A longitudinal meta-analysis on the predictive nature of early numeracy.
- Reciprocal development of numeracy, mathematical language, and general language skills from age 2 ½ to 4 years: A cross-lagged panel perspective - ScienceDirect


